How many items per column?

Three to five. More than that and you won't have time to discuss them, which means the meeting produced a list and not a decision. If you have ten Stops, dot-vote and pick one. The other nine will come back next retro, and that's fine: a thing that survives two retros without dying is a thing the team genuinely cares about.

Yes. Open a doc Monday morning with the three prompts, give the team until Wednesday end-of-day to add cards, then meet for fifteen minutes Thursday to dot-vote and pick commitments. Works well for distributed teams across multiple time zones. Bonus: the people who hate retros tend to write better cards async than they speak in meetings, so you'll often get more honest input.

Should the manager facilitate?

Rotate. Engineering managers facilitating their own team's retros makes it harder for the team to raise management-shaped Stops, because the management-shaped Stops are usually about the manager. If you must facilitate, sit out the discussion and only intervene to keep time. Bring snacks to compensate.

How often should we change format?

Every four to six retros. Even good formats go stale and the team starts answering on autopilot. Switching to 4Ls or KALM for a sprint and coming back to Start, Stop, Continue refreshes both. Think of it like rotating the coffee blend: same ritual, different notes.

What about items we agreed on but didn't do?

Start the next retro by reviewing them. Two minutes, no judgement, just visibility. If the same Stop comes up three retros in a row without changing, the team has a structural problem the format can't fix, and the conversation needs to be a different conversation, with different people in the room.

The toolkit · Start, Stop, Continue

Start, Stop, Continue retrospective

The classic three-column retro. Pick a fresh prompt for each column and run a 30-minute retro your team will actually finish on time.

30 prompts · 3 columns · free

▶️

Start

What's one thing we should start doing next sprint?

10 prompts in this bucket

⏹️

Stop

What should we stop doing immediately?

10 prompts in this bucket

✅

Continue

What's working that we should protect?

10 prompts in this bucket

Try another framework

Frameworks that pair well with Start, Stop, Continue.

4Ls

Liked, Learned, Lacked, Longed for. More texture than 3-column.

DAKI

Drop, Add, Keep, Improve. Action-first, easy to translate to a backlog.

KALM

Keep, Add, Less, More. Subtler than Start/Stop/Continue.

Or see all 10 retro frameworks.

About the Start, Stop, Continue retrospective

Start, Stop, Continue is the format every team eventually returns to, even after flirting with sailboats, gardens, and sad pigs. Three columns, three questions, no warm-up. It works on a team running its first retro and on a team that has been doing them since 2014. The format leans toward action: every Start and Stop is implicitly a commitment, which is why it's the format of choice for teams sick of retros that produce sticky notes and not change. The Continue column is the quiet hero. It's where the team protects the rituals that are working, before some new manager arrives and decides to 'optimise' them. Use Start, Stop, Continue when the team needs to decide something. The team will also need to talk about feelings sometimes, but that's what Mad, Sad, Glad and 4Ls are for.

How long should a Start, Stop, Continue retro take?

Thirty minutes for a two-week sprint. Five minutes of context, twelve minutes of silent writing across the three columns, twelve minutes of discussion, one minute to commit. For a one-week sprint, halve it. If your retro keeps running over, the issue is almost never the format. It's that the team is using the meeting to do work that should have happened during the sprint, like a deferred argument finally getting its day in court.

Three ways to ruin a Start, Stop, Continue retro

First: skipping silent writing. The room defaults to whoever speaks first, the meeting becomes one person's retro. Second: leaving Continue empty. If everyone has Stops and nobody has Continues, the team is reading itself as failing when it isn't, and the Stops will overcorrect into a different kind of broken. Third: walking out without owners. A Stop nobody owns will be a Stop again next sprint, and the sprint after, and the sprint after that, until somebody finally writes it on a wall in marker.

Start, Stop, Continue versus DAKI

DAKI (Drop, Add, Keep, Improve) is Start, Stop, Continue with an extra Improve column for things that are nearly working. Use DAKI when the team is mostly happy and the change you need is fine-tuning, not direction. Use Start, Stop, Continue when something has to change at the level of what the team does, not how well they do it. The honest test: can you finish the sentence 'this would be fine if only we…'? If yes, DAKI. If no, you're in Start, Stop, Continue territory.

What if nobody contributes?

Start the meeting silent and written. Three minutes per column, no talking, everyone types into a shared doc. The quietest people on the team get a fair shot, the meeting starts with content instead of a void, and the loudest voice doesn't get to set the frame for everyone else. Run this for three sprints in a row before deciding it doesn't work. If after three sprints the doc is still empty, the issue isn't the format, it's that the team has stopped trusting that anything they say will matter, and that's a different conversation with different people in the room.

Frequently asked

How many items per column?: Three to five. More than that and you won't have time to discuss them, which means the meeting produced a list and not a decision. If you have ten Stops, dot-vote and pick one. The other nine will come back next retro, and that's fine: a thing that survives two retros without dying is a thing the team genuinely cares about.
Can we run it async?: Yes. Open a doc Monday morning with the three prompts, give the team until Wednesday end-of-day to add cards, then meet for fifteen minutes Thursday to dot-vote and pick commitments. Works well for distributed teams across multiple time zones. Bonus: the people who hate retros tend to write better cards async than they speak in meetings, so you'll often get more honest input.
Should the manager facilitate?: Rotate. Engineering managers facilitating their own team's retros makes it harder for the team to raise management-shaped Stops, because the management-shaped Stops are usually about the manager. If you must facilitate, sit out the discussion and only intervene to keep time. Bring snacks to compensate.
How often should we change format?: Every four to six retros. Even good formats go stale and the team starts answering on autopilot. Switching to 4Ls or KALM for a sprint and coming back to Start, Stop, Continue refreshes both. Think of it like rotating the coffee blend: same ritual, different notes.
What about items we agreed on but didn't do?: Start the next retro by reviewing them. Two minutes, no judgement, just visibility. If the same Stop comes up three retros in a row without changing, the team has a structural problem the format can't fix, and the conversation needs to be a different conversation, with different people in the room.

When not to use it

Conflict-recovery or post-incident retros. Start, Stop, Continue is designed for steady-state teams making fine adjustments, not for teams that are still raw from last sprint. After something painful, run Mad, Sad, Glad first to surface what people are actually feeling. Come back to Start, Stop, Continue next sprint to commit to changes.

How to run a Start, Stop, Continue retro

1Block 30 minutes. Two-week sprint, 30 minutes is enough. One-week sprint, 20 is plenty.
2Open with a 60-second context recap. What happened in the sprint, in plain language.
3Show the three prompts (one for Start, one for Stop, one for Continue). Give everyone four minutes of silent writing against each.
4Round-robin discussion, four to five minutes per column. Loudest voices go last.
5Vote on top items. Dot-vote, two votes per person, only on Start and Stop.
6Pick one Start and one Stop the team will commit to. Assign an owner for each. Continue items go in the team handbook.
7Send the deck and the two commitments in Slack within an hour. Otherwise the meeting didn't happen.